j.1467-9701.2009.01184.x[1](10)

人民币 汇率 升值 经济影响 进出口贸易

EXPORTPERFORMANCEINEASTANDSOUTHEASTASIA29

Inlinewiththestandardpracticeintimeserieseconometrics,thetimeseriespropertyofdatawastestedattheoutsetusingtheaugmentedDickey–Fuller(ADF)test.Accordingtothetestresults,thevariablesunderconsiderationdonothavethesameorderofintegrationineachcountry.Forexample,inthePRC,therealexchangerateisstationary(I(0))whileothervariablesarenon-stationary(I(1));inThailand,totalexports,exportsinSITC7,realexchangerateandgeneratedproductioncapacityarestationarywhileothersarenon-stationary.

Underthedifferentorderofintegration,thefashionablecointegrationeconometricprocedures,suchasthetwo-stepresidual-basedprocedureadoptedbyEngleandGranger(1987),andthesystem-basedreducedrankregressionapproachofJohansen(1988,1991)formodellingnon-stationarydataareinap-propriate.TheeconometricanalysisinthisstudyisbasedontheGSMprocedure(Hendryetal.,1984;WickensandBreusch,1988;Hendry,1995;Pesaranetal.,2001).TheGSMprocedureisapplicablewhenasetofvariablesincludesaseriesthatisnon-stationaryoramixtureofnon-stationaryandstationary.Inthecaseofa nitesampleandnon-stationarydataseries,thisproceduretendstoprovideamorepreciseestimatethantheJohansenprocedure.Inparticular,theJohansenproceduretendstodeterioratesigni cantlyinsmallsamples,generatingestimateswith‘fattails’(frequentoutliers)andsometimessubstantialmeanbias(Hargreaves,1994).TheGSMcanbewrittenintermsofshort-runandlong-run(cointegration)relationshipsasinequation

(2):

ð2Þ

whereaisaconstant,Ytistheendogenousvariable,

Xj,tisthejthexplanatoryvariableandAiandBijaretheparameters.

andthelong-runthe1

isgivenbyC0C1.

Equation(2)istheparticularformulationgenerallyusedasthe‘maintainedhypothesis’ofthespeci cationsearch.Theestimationprocedureinvolves rstestimatingtheunrestrictedequation(2),andthenprogressivelysimplifyingitbyrestrictingstatisticallyinsigni cantcoef cientstozeroandreformulatingthelagpatternswhereappropriateintermsoflevelsanddifferencestoachieveorthogonality.Aspartofthespeci cationsearch,itisnecessarytocheckrigorouslyateverystageeventhemoregeneralofmodelsforpossiblemisspeci cation.Suchcheckswillinvolvebothavisualexaminationoftheresidualfromthe ttedversionofthemodelandtheuseoftestsforserialcorrelation,heteroscedasticityandnormalityintheresidual,andtheappropriatenessoftheparticularfunctionalformused.Inparticular,anysuggestionofautocorrelationintheresidualshouldleadtoarethinkabouttheformofthegeneralmodel.Aboveall,theoreticalconsistencymustbeborneinmindthroughoutthetesting-downprocedure.

Ó2009TheAuthor

JournalcompilationÓBlackwellPublishingLtd.2009